CompTIA SecAI+ Practice Test

Answer choices

1. A. AI-assisted obfuscation that changes surface form while preserving behavior
2. B. Automated attack generation that creates new payloads or malware capabilities from requirements
3. C. AI-assisted social engineering that personalizes persuasive messages for selected victims
4. D. AI-assisted reconnaissance that gathers and prioritizes information about potential targets

Wrong-answer review

• B. Automated attack generation that creates new payloads or malware capabilities from requirements: Attack generation creates attack artifacts, but the described repeated rewriting specifically preserves behavior while altering appearance.
• C. AI-assisted social engineering that personalizes persuasive messages for selected victims: Social engineering manipulates people rather than mutating malware representation.
• D. AI-assisted reconnaissance that gathers and prioritizes information about potential targets: Reconnaissance gathers target information and does not describe code mutation.

Answer choices

1. A. Allow automated rollback only under approved thresholds and change controls, with full logging.
2. B. Require a human to execute every rollback manually after the AI recommends it, even when preapproved thresholds are met.
3. C. Use AI-assisted change approval before deployment but provide no automated rollback path after release.
4. D. Permit the AI agent to rewrite and redeploy any detection rule whenever it predicts lower false-positive rates.

Answer choices

1. A. Ask the model to invent missing evidence so the report is complete.
2. B. Automatically delete the underlying evidence after summarization.
3. C. Synthesize and summarize the validated incident material, then require human review.
4. D. Treat the generated draft as final without verification.

Answer choices

1. A. Model inversion
2. B. Apply provenance watermarks to generated content so recipients can identify AI-produced material
3. C. Distributed denial-of-service (DDoS)
4. D. Code linting

Answer choices

1. A. Data balancing to change class representation in the training dataset
2. B. Multi-shot prompting with several worked examples in the prompt
3. C. Pruning low-impact model parameters to reduce resource use
4. D. Model validation against a representative holdout or test dataset

Answer choices

1. A. Provide analysts a raw model artifact and require them to build their own interaction interface
2. B. An approved security chatbot connected to authorized data sources
3. C. A static firewall rule list
4. D. An offline checksum utility

Answer choices

1. A. Retrain the model on the production request stream before controls are checked.
2. B. Eliminate the need for post-deployment monitoring.
3. C. Replace data lineage records with deployment logs.
4. D. Validate that the deployed implementation matches approved security and operational requirements.

Answer choices

1. A. The European Union Artificial Intelligence Act, a binding legal framework applying to covered AI uses in the EU
2. B. ISO/IEC 42001:2023, an international standard for an organizational AI management system
3. C. OECD AI Principles
4. D. The NIST AI Risk Management Framework, a voluntary framework for managing AI risk and trustworthiness

Answer choices

1. A. Model theft
2. B. Jailbreaking
3. C. Use data poisoning to modify training examples and influence future model behavior
4. D. Membership inference

Answer choices

1. A. Rebalance the training data so minority attack examples receive appropriate representation.
2. B. Increase prompt token limits so each analyst query can include more historical event context
3. C. Encrypt the model weights at rest.
4. D. Replace the validation set with the training set.